Quote from drp7804:
Maybe the definition of "tick" in this case depends on the type of data feed. If you're getting full market depth (something like TotalView-ITCH), then the ticks are basically individual orders w/ instructions to add or delete from your local book. If it's just top-of-book, then that's potentially any update to the inside bid/ask price/size, as well as last trade information (ie the "ticks" would be less frequent).
Even if all you really cared about was last trade data, that can still be very bursty at times and you'll probably have many occasions where multiple trades (maybe tens, hundreds[?]) take place at the exchange before you have a chance to react. Sure, you'd be able to read-in each of those trades after the fact and still maintain your processing timeline (due to buffering in the socket handler, etc)... but it's probably unrealistic to assume you'd be able to react to each and every trade <i><u>before</u></i> the next one takes place at the exchange. I'm doubting this is what you were thinking, but thought it was at least worth mentioning.
Maybe put a better way, input buffering would allow you to handle bursts of ticks separated by microseconds and still keep your timeline over the long-term, but your reaction time to any given tick (ability to manage your orders) might be on the level of some number of milliseconds.